0
$\begingroup$

For my master thesis I am studying the effect of certain individual characteristics, such as financial literacy, on the probability of using a fintech. My dependent variable is a continuous variable between 0 and 1, as respondents were ask to self asses what the probability of them adopting a new technology would be.

I have run my fractional outcomes' regression (fracreg logit) on Stata and calculated the marginal affects (dydx). I am now writing my results section and a few questions have come up. I have researched online and from what I can tell there is not a lot on fractional outcomes regression that has been simplified enough for laymen to read.

My questions are the following:

  1. In a fractional outcomes logistic regression, do my independent variables have to be transformed to be between 0 and 1 before including them in the regression? - I have found mixed indications of this online

  2. How do I check if my fractional regression is robust? I have read about Quasi-Maximum likelihood estimation test but I'm not sure if it is correct to test the robustness, or if there are other methods.

  3. Are there any assumptions or steps in my methodology that I should include? I have checked variables for multicollinearity, any other important tests or steps in my process I should look at?

I have research quite a bit and but since it seems like Fractional Outcomes regression is not as established information is mixed and not always clear.

$\endgroup$
1
$\begingroup$

There are some similar questions on this site, so you would benefit from a look:

But in short:

  1. In a fractional outcomes logistic regression, do my independent variables have to be transformed to be between 0 and 1 before including them in the regression? - I have found mixed indications of this online

NO, they need not. Avoid information sources which tell you otherwise!

  1. How do I check if my fractional regression is robust? I have read about Quasi-Maximum likelihood estimation test but I'm not sure if it is correct to test the robustness, or if there are other methods.

Since your response is not binomial, you must use quasilikelihood. In R that would be glm with family=quasibinomial; other software should have similar options. Alternatively you could look into robust standard errors.

  1. Are there any assumptions or steps in my methodology that I should include? I have checked variables for multicollinearity, any other important tests or steps in my process I should look at?

Mostly, it's the same as for any other glm, so see Diagnostics for logistic regression?. Make a lot of plots.

On CRAN there is an R package frm for these models. One if its authors have a dedicated web page with many references.

An alternative could be beta regression.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.